Stockfish Testing Queue

Finished - 188 tests

19-02-06 Ala 8x8KingPSQTune2 diff
28739/30000 iterations
59714/60000 games played
60000 @ 60+0.6 th 1 STC with raw unsmoothed values did +1.5 elo. Finish tune on rebased master with lower variance to help convergence and some pawn value co-tuning.
19-02-07 Ala PawnPushTweak diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 18211 W: 4023 L: 4097 D: 10091
sprt @ 10+0.1 th 1 Take 3, try to avoid "blocking" a pawn push threat by doing a reckless counter pawn push. (fixed bench)
19-02-06 Ala 8x8KingPSQT diff
LLR: -1.68 (-2.94,2.94) [0.50,4.50]
Total: 79475 W: 17620 L: 17321 D: 44534
sprt @ 10+0.1 th 1 Test 106K 8x8 king PSQT tune results (non-smoothed)
19-02-06 Ala PawnPushTweak diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 23652 W: 5218 L: 5265 D: 13169
sprt @ 10+0.1 th 1 Take 2 : fix conditions so that double square pawn push doesn't require intermediate square to be safe, only not attacked by a pawn.
19-02-06 Ala PawnPushTweak diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 13431 W: 2882 L: 2980 D: 7569
sprt @ 10+0.1 th 1 Try to better assess pawn push threats to better evaluate positions like those in the lost game 11 of TCEC SuFi
19-02-05 Ala KingSafetyParams diff
LLR: -2.96 (-2.94,2.94) [0.00,3.50]
Total: 88935 W: 14711 L: 14640 D: 59584
sprt @ 60+0.6 th 1 LTC for king safety params tune (yellow STC was with elo gainer bounds, would have passed with param tweak bounds)
19-02-06 Ala DynamContemptTweak diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 10501 W: 2291 L: 2404 D: 5806
sprt @ 10+0.1 th 1 Try to make SF accept smaller compensation to simplify if position is bad (< -0.4cp against itself) to limit "contempt blunders".
19-02-05 Ala KingSafetyParams diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 88849 W: 19567 L: 19294 D: 49988
sprt @ 10+0.1 th 1 Test tuning results
19-02-04 Ala KingSafetyTune2 diff
24023/25000 iterations
49975/50000 games played
50000 @ 60+0.6 th 1 Finish king safety tuning with tweaked variances.
19-01-28 Ala 8x8KingPSQTune diff
51225/75000 iterations
106500/150000 games played
150000 @ 60+0.6 th 1 Retry the 8x8 king PSQT idea. Tune starting from Kurt values.
19-02-03 Ala master diff
ELO: 17.71 +-1.8 (95%) LOS: 100.0%
Total: 40000 W: 6661 L: 4624 D: 28715
40000 @ 60+0.6 th 1 Regression/progression test against SF10 after "Less king danger if we have a knight near by to defend it." of February, 3rd.
19-02-02 Ala KingSafetyTune diff
36266/60000 iterations
76026/120000 games played
120000 @ 60+0.6 th 1 Tune king safety, including knight in king ring bonus, with more variance for newly added/changed values.
19-01-29 Ala PawnAttMob diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 21091 W: 4633 L: 4693 D: 11765
sprt @ 10+0.1 th 1 Test tune results
19-01-28 Ala PawnAttMobTune diff
72039/75000 iterations
149556/150000 games played
150000 @ 30+0.3 th 1 Score differently mobility depending on how many squares would have been in the mobility area if not for enemy pawn attacks.
19-01-28 Ala KBPsKPs diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 14980 W: 3237 L: 3327 D: 8416
sprt @ 10+0.1 th 1 Bench do not change but this patch do improve some draw detection rules in addition to some renaming and comments for clarity, though I doubt t makes a big enough difference to pass SPRT.
19-01-26 Ala pinners2 diff
LLR: 3.40 (-2.94,2.94) [0.00,3.50]
Total: 140285 W: 23416 L: 22825 D: 94044
sprt @ 60+0.6 th 1 LTC for xoroshiro
19-01-25 Ala MultiImbalance diff
LLR: -1.52 (-2.94,2.94) [0.00,3.50]
Total: 41359 W: 6815 L: 6791 D: 27753
sprt @ 60+0.6 th 1 STC results weren't good, but I suspect this may scale and framework is empty. Low TP.
19-01-25 Ala BishopPawnPassersAlt diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 22700 W: 4902 L: 4955 D: 12843
sprt @ 10+0.1 th 1 Take 1
19-01-25 Ala MultiImbalanceHalf diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 36525 W: 7930 L: 7915 D: 20680
sprt @ 10+0.1 th 1 Test with half effect. I don't expect this to pass, but it should be an interesting data point.
19-01-25 Ala BishopPawnPassersAlt diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 12827 W: 2763 L: 2864 D: 7200
sprt @ 10+0.1 th 1 Take 2
19-01-25 Ala MultiImbalance diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 17334 W: 3763 L: 3842 D: 9729
sprt @ 10+0.1 th 1 Test tune results
19-01-23 Ala MultiImbalanceTune diff
71986/75000 iterations
150000/150000 games played
150000 @ 20+0.2 th 1 Starting from smoothed values of previous tune, continue the tuning with better adapted range for variables, smaller variance (the previous one was great to get a quick approximation starting from all 0 but was too sensitive to converge). Also, as the subcases in 3Mv2R and 3MvQ proved important, add subcases for 2RvQ and MRvQ
19-01-24 Ala LongDiagBishTweak diff
LLR: -0.19 (-2.94,2.94) [0.50,4.50]
Total: 335 W: 67 L: 76 D: 192
sprt @ 10+0.1 th 1 Also apply long diagonal bishop bonus when the bishop is on one of the two center squares.
19-01-24 Ala UnpinnedAttackBB diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 11399 W: 2447 L: 2555 D: 6397
sprt @ 10+0.1 th 1 Previous tests were invalid because of a subtle bug (the ordering in bitboard & square matter). Try again with a rebase to latest master. Comparison against non-functional version to gauge eval gain, the perf hit is too heavy for now.
19-01-24 Ala UnpinnedAttackBB diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 9169 W: 1944 L: 2063 D: 5162
sprt @ 10+0.1 th 1 Try again bigger bonus difference with bugfix included.
19-01-23 Ala MultiImbalance diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 18199 W: 3961 L: 4036 D: 10202
sprt @ 10+0.1 th 1 Use early tune values to gauge progress.
19-01-23 Ala TrappedRookMix diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 18281 W: 4003 L: 4077 D: 10201
sprt @ 10+0.1 th 1 Increase bonus when no castling right, but less so when rook and king on different ranks.
19-01-23 Ala MultiImbalanceTune diff
30534/75000 iterations
63891/150000 games played
150000 @ 20+0.2 th 1 The quadratic imbalance handles how two piece type interact but not specific imbalances. There are some imbalance scoring patch attempt. This tune is for a much more generalized version. Starting values at 0 with high variance to hopefully not get stuck.
19-01-22 Ala KRBKRB3 diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 14373 W: 3104 L: 3197 D: 8072
sprt @ 10+0.1 th 1 Take 3 : try a quadratic bonus
19-01-22 Ala KRBKRB2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 16345 W: 3541 L: 3625 D: 9179
sprt @ 10+0.1 th 1 Take 2 : double bonus
19-01-22 Ala KRBKRB diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 19407 W: 4185 L: 4254 D: 10968
sprt @ 10+0.1 th 1 Take 1 : bonus for king closer to pawns in rook+bishop endgames, scaled with distance and pawn relative rank
19-01-22 Ala UnpinnedAttackBB diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 21815 W: 4802 L: 4859 D: 12154
sprt @ 10+0.1 th 1 Take 2 with different parameters balance. Testing against version with the new bb overhead to first find something improving eval and try and get to have several places with such improvement to overcome overhead.
19-01-21 Ala UnpinnedAttackBB diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 14986 W: 3255 L: 3345 D: 8386
sprt @ 10+0.1 th 1 Test against non-functional version with same slowdown to evaluate if this approach deserves more work (e.g. optimization, fixing some accuracy limitations, using the new bitboards in other places thus making the perf cost more bearable).
19-01-21 Ala UnpinnedAttackBB diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 8102 W: 1731 L: 1856 D: 4515
sprt @ 10+0.1 th 1 Take 1 : reduce hanging bonus if the threatening piece is pinned to a queen.
19-01-21 Ala DiagonalPromoBishop diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15207 W: 3316 L: 3405 D: 8486
sprt @ 10+0.1 th 1 With 31m's fix
19-01-21 Ala DiagonalPromoBishop diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 19372 W: 4186 L: 4255 D: 10931
sprt @ 10+0.1 th 1 Take 2 fixed : Half-bonus + correct own passer file requirement for white
19-01-21 Ala DiagonalPromoBishop diff
LLR: 0.10 (-2.94,2.94) [0.50,4.50]
Total: 4985 W: 1092 L: 1062 D: 2831
sprt @ 10+0.1 th 1 Take 2 : half bonus.
19-01-21 Ala DiagonalPromoBishop diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 12403 W: 2706 L: 2809 D: 6888
sprt @ 10+0.1 th 1 Endgame bonus for long diagonal bishop with favorable pawn setup
19-01-20 Ala RestrictQueenPinnedAtta diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 5932 W: 1286 L: 1422 D: 3224
sprt @ 10+0.1 th 1 Inspired by this Bryan analysis : https://groups.google.com/forum/#!topic/fishcooking/LHkYy572huI First attempt with still some clear limitations (see code comments).
19-01-19 Ala ScaleRookPawn diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 21106 W: 4584 L: 4644 D: 11878
sprt @ 10+0.1 th 1 Take 2 with additional simplifications
19-01-19 Ala ScaleRookPawn diff
LLR: -0.04 (-2.94,2.94) [0.50,4.50]
Total: 22 W: 3 L: 5 D: 14
sprt @ 10+0.1 th 1 Take 2 : fix a bug, optimize a bit, and tweak
19-01-19 Ala ScaleRookPawn diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 51788 W: 11446 L: 11355 D: 28987
sprt @ 10+0.1 th 1 First unoptimized attempt at scaling down eval in some typically drawn rook-pawn endgames
19-01-19 Ala TweakFailTM diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 1708 W: 314 L: 471 D: 923
sprt @ 10+0.1 th 1 Use results of tuning
19-01-18 Ala TweakFailTMTune diff
45705/50000 iterations
94965/100000 games played
100000 @ 30+0.3 th 1 The previous tune variations showed that two eval terms can likely be simplified away. Remove them, compensate the removal in other values, and continue the tune from there.
19-01-18 Ala TweakFailTMTune diff
20139/40000 iterations
41799/80000 games played
80000 @ 20+0.2 th 1 Build upon the previous tune
19-01-18 Ala TweakFailTMTune diff
15241/37500 iterations
31588/75000 games played
75000 @ 20+0.2 th 1 Tune for a more general formula based on the idea behind my TweakFailTM patch about to fail yellow
19-01-17 Ala TweakFailTM diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 65792 W: 14629 L: 14468 D: 36695
sprt @ 10+0.1 th 1 A cp with eval close to 0 means more than a cp with high absolute eval, try and tweak TM to account for it. No functional change.
19-01-14 Ala EdgeMobMinor diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 40642 W: 8948 L: 8912 D: 22782
sprt @ 10+0.1 th 1 Take 2 (stronger bonus as first test was neutral) with Viz optimization
19-01-13 Ala EdgeMobMinor diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 20274 W: 4396 L: 4460 D: 11418
sprt @ 10+0.1 th 1 If low mobility minor and only mobility squares are on the edges, apply a small penalty
19-01-13 Ala EdgeMobMinor diff
LLR: -0.00 (-2.94,2.94) [0.50,4.50]
Total: 25 W: 5 L: 5 D: 15
sprt @ 10+0.1 th 1 Take 2 - test a stronger malus